# Self-Organizing Maps with Optimized Latent Positions

Source: [Self-Organizing Maps with Optimized Latent Positions](<https://www.sota2.com/research/paper/69e06cf77d0172140ec7c933>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-17T05:00:04.441Z

Authors: Seiki Ubukata, Akira Notsu, Katsuhiro Honda

Published: 2026-04-15T08:42:49.000Z

## Abstract

Self-Organizing Maps (SOM) are a classical method for unsupervised learning, vector quantization, and topographic mapping of high-dimensional data. However, existing SOM formulations often involve a trade-off between computational efficiency and a clearly defined optimization objective. Objective-based variants such as Soft Topographic Vector Quantization (STVQ) provide a principled formulation, but their neighborhood-coupled computations become expensive as the number of latent nodes increases. In this paper, we propose Self-Organizing Maps with Optimized Latent Positions (SOM-OLP), an objective-based topographic mapping method that introduces a continuous latent position for each data point. Starting from the neighborhood distortion of STVQ, we construct a separable surrogate local cost based on its local quadratic structure and formulate an entropy-regularized objective based on it. This yields a simple block coordinate descent scheme with closed-form updates for assignment probabilities, latent positions, and reference vectors, while guaranteeing monotonic non-increase of the objective and retaining linear per-iteration complexity in the numbers of data points and latent nodes. Experiments on a synthetic saddle manifold, scalability studies on the Digits and MNIST datasets, and 16 benchmark datasets show that SOM-OLP achieves competitive neighborhood preservation and quantization performance, favorable scalability for large numbers of latent nodes and large datasets, and the best average rank among the compared methods on the benchmark datasets.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Manifold Learning on Iris](<https://www.sota2.com/research/sota/manifold-learning-on-iris.md>): Manifold Learning; Iris; Average (TW + CN): 0.9869.

- [Manifold Learning on Seeds](<https://www.sota2.com/research/sota/manifold-learning-on-seeds.md>): Manifold Learning; SEEDS; TW + CN / 2 Score: 98.06.

- [Manifold Learning on Heart Disease](<https://www.sota2.com/research/sota/manifold-learning-on-heart-disease.md>): Manifold Learning; HEART DISEASE; (TW + CN)/2: 93.5.

- [Manifold Learning on Ionosphere](<https://www.sota2.com/research/sota/manifold-learning-on-ionosphere.md>): Manifold Learning; ionosphere; TW+CN Metric: 0.9151.

- [Manifold Learning on Breast Cancer (WDBC)](<https://www.sota2.com/research/sota/manifold-learning-on-breast-cancer-wdbc.md>): Manifold Learning; Breast Cancer (WDBC); (TW + CN)/2: 0.9517.

- [Manifold Learning on Banknote Authentication](<https://www.sota2.com/research/sota/manifold-learning-on-banknote-authentication.md>): Manifold Learning; banknote-authentication; (TW + CN)/2: 99.63.

- [Manifold Learning on Yeast ( (TW + CN)/2 )](<https://www.sota2.com/research/sota/manifold-learning-on-yeast-tw-plus-cn-2.md>): Manifold Learning; Yeast; (TW + CN)/2: 0.96.

- [Manifold Learning on Rice Cammeo and Osmancik](<https://www.sota2.com/research/sota/manifold-learning-on-rice-cammeo-and-osmancik.md>): Manifold Learning; Rice Cammeo and Osmancik; (TW + CN)/2: 99.11.

- [Manifold Learning on Dry Bean](<https://www.sota2.com/research/sota/manifold-learning-on-dry-bean.md>): Manifold Learning; Dry Bean; (TW + CN)/2: 98.65.

- [Manifold Learning on Saddle Manifold](<https://www.sota2.com/research/sota/manifold-learning-on-saddle-manifold.md>): Manifold Learning; Saddle Manifold; TW: 99.86.

- [Manifold Learning on Wine](<https://www.sota2.com/research/sota/manifold-learning-on-wine.md>): Manifold Learning; Wine; (TW + CN)/2 Score: 95.93.

- [Manifold Learning on Ecoli](<https://www.sota2.com/research/sota/manifold-learning-on-ecoli.md>): Manifold Learning; ecoli; TW + CN / 2 Score: 96.71.

- [Manifold Learning on Dermatology](<https://www.sota2.com/research/sota/manifold-learning-on-dermatology.md>): Manifold Learning; Dermatology; Performance ((TW + CN)/2): 96.07.

- [Manifold Learning on Digits](<https://www.sota2.com/research/sota/manifold-learning-on-digits.md>): Manifold Learning; Digits; (TW + CN)/2: 97.66.

- [Manifold Learning on Image Segmentation](<https://www.sota2.com/research/sota/manifold-learning-on-image-segmentation.md>): Manifold Learning; Image Segmentation; Composite Score ((TW + CN)/2): 0.987.

- [Scalability analysis on Digits](<https://www.sota2.com/research/sota/scalability-analysis-on-digits.md>): Scalability Analysis; Digits; Per-iteration Wall-Clock Time (ms): 1.

- [Manifold Learning on Sonar](<https://www.sota2.com/research/sota/manifold-learning-on-sonar.md>): Manifold Learning; SONAR; Combined TW+CN Score: 93.44.

- [Manifold Learning on Glass](<https://www.sota2.com/research/sota/manifold-learning-on-glass.md>): Manifold Learning; Glass; (TW + CN)/2: 0.9497.

- [Topographic Mapping on Digits scikit-learn (train test combined)](<https://www.sota2.com/research/sota/topographic-mapping-on-digits-scikit-learn-train-test-combined.md>): Topographic Mapping; Digits scikit-learn (train test combined); TW: 0.9864.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.13622>)

[PDF](<https://arxiv.org/pdf/2604.13622.pdf>)
